pydantic-ai 2.5.0

v2.5.0 (2026-07-03)

on Python PyPI

What's Changed

🚀 Features

• Add sanitize_messages for inbound message-history hardening by @dsfaccini in #6169
• Round-trip multimodal tool returns through the AG-UI and Vercel AI adapters (history and streaming) by @dsfaccini in #5255

🐛 Bug Fixes

• Document why _clean_message_history merges consecutive ModelRequests regardless of conversation_id/metadata by @adtyavrdhn in #6221
• Align LLMJudge _build_prompt section order with its few-shot examples by @immuhammadfurqan in #6111
• Reject duplicate deferred tool call IDs by @hramezani in #6241
• Fix MoonshotAI Kimi thinking support and add models to KnownModelNames by @SuperMarioYL in #6208
• fix(dbos,prefect): preserve runtime toolsets in durable runs by @Trollgeir in #6076
• Exclude first-class token counts from gen_ai.usage.details.* OTel attributes to fix double-counting by @adtyavrdhn in #6244
• fix(dbos,prefect): preserve runtime event stream handlers by @Trollgeir in #6093
• Reject non-object agent spec inputs by @Ghraven in #5820
• Run the sync streaming wrappers on an anyio portal thread; fix async stream_text early-break teardown by @DouweM in #6199
• fix(bedrock): only synthesize a leading user message for models that require it by @DouweM in #6255
• fix(bedrock): correct bedrock_supported_media_kinds_in_tool_returns for Meta, Mistral, and Qwen by @DouweM in #6254

📦 Dependencies

• Avoid Python 3.14 numpy source builds in examples by @hramezani in #6246

New Contributors

• @pragnyanramtha made their first contribution in #5490
• @Sehlani042 made their first contribution in #6200
• @ChunkyTortoise made their first contribution in #5739

Full Changelog: v2.4.0...v2.5.0